Artificial Neural Network Based Automatic Modulation Classification System Applied to FPGA

Abstract

The wireless communication systems face rapid growth, driven by advances in new technologies such as the 5G and the Internet of Things. However, this growth faces a limitation: the scarcity of frequencies in the electromagnetic spectrum, demanding efficient technologies to improve its utilization. For this reason, this work aimed to construct an Automatic Modulation Classification system and implement it in both software and hardware, using an FPGA. The resulting models can classify five modulations and a noise-only signal, using an Artificial Neural Network architecture, which was constructed based on the test of over 2000 different topologies, resulting in distinct configurations for each technology due to their intrinsic limitations. Both setups achieved approximately 90% of accuracy when the SNR is ≥4 dB and are capable of outperforming similar works developed so far, as it uses a set of inputs that require less computational time and resource utilization on its execution.